Research On Pavement Damage Detection Method Based On UAV

With the significant increase in highway maintenance mileage in recent years,the industry’s demand for accuracy in road damage detection has also gradually risen.The current highway inspection strategies have become inadequate to meet the detection needs of our country’s highways.In response to the problems of low detection efficiency and low accuracy of the existing automatic detection methods and easy to cause traffic jams,a damage detection method based on aerial photography of road surface by UAV is proposed.Firstly,the specific condition of existing road damage is analyzed,and a refined labeling dataset containing different cracks,potholes and repairs in six categories of road damage is constructed for the problems of incomprehensive labeling categories,low pixel resolution,random damage markings and unrealizable segmentation tasks in the existing road damage dataset.With YOLOv5 s as the base network,CBAM is introduced in the CSP module of the network structure,and an improved YOLOv5s-based road damage detection model is proposed to accomplish the common pavement damage classification and detection tasks.Image pre-processing and damage texture feature analysis are required before highway pavement image detection to extract its grayscale co-occurrence matrix parameters,and the obtained grayscale co-occurrence matrix is used as one of the input parameters of the model.The experiments prove that the detection accuracy of the improved YOLOv5s-based pavement damage classification model is high and effective,with an average detection accuracy of 91.6%.The U-Net network model is adopted as the basic framework,and the first 13 convolutional layers of VGG16 are used instead of the original downsampling part,while the upsampling part adopts bilinear interpolation instead of the inverse convolution in the original structure to construct the VGG16＿U-Net road damage semantic segmentation model to complete the multi-category pixel-level segmentation task of common road damage.The obtained segmented images are also further analyzed in the connected domain to calculate the pixel size of the breakage,and an eight-directional orthogonal skeleton algorithm is proposed for crack width calculation to reduce the computational time consumption and computational effort.The experiment proves that the VGG16＿U-Net semantic segmentation model of road damage can not only complete the visualization output of segmentation results,but also directly obtain the specific physical dimensions of pavement damage.After the field measurement comparison,the constructed segmentation model image detection accuracy is much higher than the industry detection standard and meets the industry demand.Finally,the road damage detection model based on improved YOLOv5 s and VGG16＿U-Net road damage semantic segmentation model are integrated and added with road damage pinpointing and damage condition assessment to build an interactive UAV aerial photography road damage detection system and realize the integrated detection of road damage.